Logistics park vehicle pick-up scheduling method based on adaptive discrete particle swarm algorithm

Abstract

The logistics park vehicle pickup scheduling method based on the adaptive discrete particle swarm algorithm has the following specific steps: analyzing the operation data related to the logistics park vehicle, performing K-means clustering on the target goods based on the time of each vehicle entering the park and the time of each vehicle leaving the park and the target pickup goods corresponding to each vehicle; encoding the yard allocation part; encoding the pickup goods part; setting the global search to determine the initial population; setting the local search to determine the initial population; setting the random search to determine the initial population; setting the maximum iteration number of the algorithm as iter; initializing the state set and the action set in the reinforcement learning, selecting the action by using the greedy strategy, and calculating the reward according to the state and the fitness value of the chromosome of the population at the current time. The logistics park vehicle pickup scheduling method based on the adaptive discrete particle swarm algorithm has the following beneficial effects: the waiting time of the vehicle in the pickup process can be significantly reduced, the logistics throughput is improved, and the maximum operation time of the yard is indirectly shortened to meet the demand of the logistics park yard for work efficiency.

Description

Technical Field

[0001] This invention relates to the field of vehicle scheduling technology, and specifically to a logistics park vehicle pickup scheduling method based on an adaptive discrete particle swarm optimization algorithm. Background Technology

[0002] As logistics parks expand, higher demands are placed on the efficiency of the logistics system. Efficient vehicle scheduling can significantly reduce vehicle waiting time during the pickup process and increase logistics throughput.

[0003] Current vehicle dispatching relies primarily on manual experience, which is inefficient and prone to causing traffic congestion when multiple vehicles are operating simultaneously in different parking lots. Optimizing the dispatching mechanism using relevant data to meet the operational needs of both vehicles and parking lots presents a significant challenge.

[0004] Traditional vehicle entry queue management largely relies on manual scheduling, arranging entry order according to a first-come, first-served principle. Due to a lack of scientific optimization, this leads to problems such as scheduling chaos and low customer satisfaction. Traditional metaheuristic algorithms cannot dynamically adjust key parameters, resulting in lower-than-expected solution efficiency and speed. This invention addresses the vehicle yard scheduling problem based on an adaptive discrete particle swarm optimization algorithm, using the SARSA algorithm and Q-learning from reinforcement learning to optimize parameters. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings and defects of existing technologies by providing a logistics park vehicle pickup scheduling method based on an adaptive discrete particle swarm optimization algorithm. This method can significantly reduce vehicle waiting time during pickup, increase logistics throughput, and indirectly shorten the maximum operating time of the yard to meet the efficiency requirements of logistics park yards.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a logistics park vehicle pickup scheduling method based on adaptive discrete particle swarm optimization algorithm, the specific steps of which are as follows: Analyze the operation data related to logistics park vehicles; based on the entry and exit times of each vehicle and the target pickup goods corresponding to the vehicle, perform K-means clustering on the target goods; encode the yard allocation part; encode the pickup goods part; set up a global search to determine the initial population; set up a local search to determine the initial population; set up a random search to determine the initial population; set the maximum number of algorithm iterations to iter; initialize the state set and action set in reinforcement learning, select actions using a greedy strategy, and calculate the reward based on the state and fitness value of the population chromosome at the current time; update the Q-table using the Sarsa algorithm, and determine the current Sarsa update. If the number of iterations exceeds the number of transformations, and if so, the QLearning algorithm is started to update the Q-table, and a greedy strategy is used to select actions. In each iteration, the quality of the current population is constantly assessed. Based on the trained Q-table, the discrete particle swarm parameter ω is adaptively updated using the current population state of the chromosomes, selecting the ω most favorable to the current population chromosomes. A random number between [0,1] is set, and its magnitude is compared with ω, employing both cargo selection mutation and yard selection mutation methods. The POX chromosome crossover method is selected, performing crossovers on the yard chromosomes and cargo chromosomes respectively. A two-point crossover method is selected, performing crossovers on the yard chromosomes and cargo chromosomes respectively. Chromosome update operations are performed, and the fitness values ​​of each particle in the current population are evaluated and calculated. The optimal P-value for updating the individual position is determined by comparison. i and the global optimal position P g Check if the iteration count meets the condition. If it does, stop the iteration and output the global optimal position P. g This is used as the optimal scheduling result.

[0007] As further illustrative of this invention: the analysis of logistics park vehicle-related operational data specifically involves: analyzing the historical dwell time of vehicles within the yard based on the yard's mixed-stacking rules, and providing the distribution T of vehicle pickup time in each yard. i .

[0008] As further illustrative of the present invention: the encoding of the yard allocation part specifically involves setting the chromosome length of the yard allocation part to the sum of the cargo types of all vehicles picking up cargo, each gene position being represented by an integer, arranged sequentially according to the vehicle and the cargo type picked up by the vehicle, and each integer representing the yard number that meets the requirements for the current cargo picked up by the vehicle.

[0009] As further illustrative of the present invention: the encoding of the goods to be picked up specifically involves setting the chromosome length of the goods to be picked up by the vehicles to the sum of the types of goods picked up by all vehicles, and encoding each gene directly with the vehicle number, the order in which the vehicle numbers appear indicates the order in which the goods were picked up by that vehicle.

[0010] As further illustrative of the present invention: the setting of global search to determine the initial population specifically involves: setting an array of length equal to the number of storage yards, where each value in the array corresponds to the working time of the corresponding storage yard; randomly selecting a vehicle i from the set of picking vehicles; starting from the first picking goods of the current vehicle i; adding the picking time of the storage yard that currently satisfies the target picking goods to the corresponding time in the array; selecting the shortest time as the current target storage yard; and updating the array; and so on, until all vehicles have finished picking goods.

[0011] As further illustrative of the present invention: the setting of local search to determine the initial population specifically involves: setting an array of length equal to the number of storage yards, where each value in the array corresponds to the working time of the corresponding storage yard; randomly selecting a vehicle i from the set of pickup vehicles; starting from the first pickup of the current vehicle i, adding the pickup time of the storage yard that currently satisfies the target pickup time to the corresponding time in the array; selecting the shortest time as the current target storage yard; when vehicle i has been selected, resetting the array to 0; selecting the next vehicle; and so on, until all vehicles have finished picking up their goods.

[0012] As further illustrative of the present invention: the setting of random search to determine the initial population specifically involves: randomly selecting a vehicle i from the set of pickup vehicles, randomly sorting the order of the target pickup goods for vehicle i, and randomly selecting a storage yard from the set of storage yards that meet the requirements of the sorted target goods, and so on, until all vehicles have finished picking up their goods.

[0013] As further illustrative of this invention: the specific steps for initializing the state set and action set in reinforcement learning, selecting actions using a greedy strategy, and calculating the reward based on the state and fitness value of the population chromosome at the current moment are as follows: Initialize the state set in reinforcement learning, setting the value range of the state set to [0.5, 1], and dividing the value range into 10 intervals as state1, state2...state10; Initialize the action set in reinforcement learning, setting the value range of the state set to [0, 1], and dividing the value range into 10 intervals as action1, action2...action10; Select actions based on the state using a greedy strategy, and calculate the reward based on the state and fitness value of the population chromosome at the current moment.

[0014] As further illustrative of this invention: the Sarsa algorithm updates the Q-table, determines whether the current Sarsa update count is greater than the transformation count, and if so, starts the QLearning algorithm to update the Q-table. The specific steps for selecting actions using a greedy strategy are as follows: set the algorithm transformation count t = 1, and the algorithm transformation count T = 25; the Sarsa algorithm updates the Q-table based on Q(s... t a t )=(1-a)Q(s t a t )+a(r t+1 +γQ(s t+1 a t+1 Update the Q-table; check if the current t is greater than the number of transformations. If it is, start the chromosome update operation; if it is less, start selecting the POX chromosome crossover method; the QLearning algorithm is based on Q(s) t a t )=(1-a)Q(s t a t )+a(r t+1 +γmax a Q(s t+1 a t+1 Update the Q table and use a greedy strategy to select actions.

[0015] As further illustrative of the present invention: the random number set between [0,1] is compared with ω, and the method of cargo mutation and yard selection mutation is specifically as follows: the chromosome part of the cargo is mutated and the yard selection part of the vehicle is mutated respectively, and one of the two mutations is selected as the chromosome mutation operation by using a roulette wheel betting method.

[0016] After adopting the above technical solution, the beneficial effects of this invention are as follows: This logistics park vehicle pickup and dispatching method has the following advantages:

[0017] 1. Utilize reinforcement learning algorithms to intelligently update ω in the discrete particle swarm optimization algorithm, thereby improving solution efficiency and speed;

[0018] 2. Dividing the state set of chromosomes and using the average fitness and diversity of the population as states reflects the overall state of the population, which is beneficial to improving the quality of the entire population and making it easier to obtain excellent individuals;

[0019] 3. Combining the Sarsa algorithm with the QLearning algorithm. Sarsa is an on-policy algorithm, which yields more conservative results and converges faster than QLearning, but its learning performance is inferior. QLearning is an off-policy algorithm, which offers greater potential and better global search capabilities compared to Sarsa, but its convergence speed is slower. Combining the two algorithms allows the algorithm to initially inherit the learning performance and convergence speed of Sarsa, while later inheriting the better optimization capabilities of QLearning.

[0020] 4. The vehicle scheduling algorithm proposed in this method can significantly reduce the waiting time of vehicles during the pickup process, increase logistics throughput, and indirectly shorten the maximum operation time of the yard to meet the efficiency requirements of the logistics park yard. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the algorithm principle of the present invention.

[0023] Figure 2 This is a schematic diagram of the algorithm flow of the present invention.

[0024] Figure 3 This is a comparison chart of the algorithm iteration results when the number of vehicles is 10 in this invention.

[0025] Figure 4 This is a comparison chart of the algorithm iteration results for vehicle 40 in this invention.

[0026] Figure 5 This is a diagram showing the scheduling results for vehicle 10 in this invention.

[0027] Figure 6 This is a diagram showing the scheduling results for vehicle number 40 in this invention. Detailed Implementation

[0028] See Figures 1-6 As shown, the technical solution adopted in this specific embodiment is as follows: The specific steps are as follows:

[0029] S1 analyzes the operational data related to vehicles in the logistics park. Based on the entry and exit times of each vehicle and the target goods to be picked up, K-means clustering is performed on the target goods. According to the yard mixing rules, the historical dwell time of vehicles in the yard is analyzed, and the distribution T of vehicle picking time in each yard is given. i .

[0030] S2, the yard allocation part is encoded. The chromosome length of the yard allocation part is set to the sum of the cargo types of all vehicles picking up cargo. Each gene position is represented by an integer and arranged in sequence according to the vehicle and the cargo type of the vehicle. Each integer represents the yard number that meets the requirements for the current cargo picked up by the vehicle.

[0031] S3, encode the goods to be picked up. Set the chromosome length of the goods to be picked up by all vehicles to the sum of the types of goods picked up. Each gene is directly encoded by the vehicle number. The order in which the vehicle numbers appear indicates the order in which the goods were picked up by that vehicle.

[0032] S4. Set up a global search to determine the initial population. Set up an array of length equal to the number of yards. The value in each array corresponds to the working time of the corresponding yard. Randomly select a vehicle i from the set of picking vehicles. Starting from the first picking goods of the current vehicle i, add the picking time of the yard that meets the target picking goods to the corresponding time in the array. Select the shortest time as the current target yard and update the array. Continue in this way until all vehicles have picked up their goods.

[0033] S5. Set up a local search to determine the initial population. Set up an array of length equal to the number of yards. The value in each array corresponds to the working time of the corresponding yard. Randomly select a vehicle i from the set of picking vehicles. Starting from the first picking goods of the current vehicle i, add the picking time of the yard that meets the target picking goods to the corresponding time in the array. Select the shortest time as the current target yard. When vehicle i has been selected, reset the array to 0 and select the next vehicle until all vehicles have picked up their goods.

[0034] S6, set up a random search to determine the initial population, randomly select a vehicle i from the set of pickup vehicles, randomly sort the order of the target pickup goods of vehicle i, and randomly select a yard from the set of yards that meet the requirements of the sorted target goods, and so on, until all vehicles have picked up their goods.

[0035] S7, set the maximum number of iterations of the algorithm to iter, where iter = 50.

[0036] S8: Initialize the state and action sets in reinforcement learning, select actions using a greedy strategy, and calculate the reward based on the current state and fitness value of the population chromosomes.

[0037] 1) Initialize the state set in reinforcement learning, set the value range of the state set to [0.5,1], and divide the value range into 10 intervals as state1, state2...state10;

[0038] 2) Initialize the action set in reinforcement learning, set the range of the state set to [0,1], and divide the range of values ​​into 10 intervals, which are action1, action2...action10;

[0039] 3) Use a greedy strategy to select actions based on the state, and calculate the reward based on the state and fitness of the population chromosome at the current moment.

[0040] S9, the Sarsa algorithm updates the Q-table, checking if the current Sarsa update count is greater than the transformation count. If it is, the QLearning algorithm starts updating the Q-table, using a greedy strategy to select actions.

[0041] 1) Set the number of algorithm transformations t = 1, and the total number of algorithm transformations T = 25;

[0042] 2) The Sarsa algorithm is based on Q(s) t a t )=(1-a)Q(s t a t )+a(r t+1 +γQ(s t+1 a t+1 Update the Q table, where Q(s) t a t ) represents state s t Use action a t The value, where 'a' represents the soft update weight, and 'r' represents the value. t+1 Let γ represent the reward at time step t+1, and γ represent the discount rate, where t = t+1

[0043] 3) Determine if the current t is greater than the number of conversions. If it is, start S14; if it is less, start S12.

[0044] 4) The Qkearning algorithm is based on Q(s) t a t )=(1-a)Q(s t a t )+a(r t+1 +γmaxa Q(s t+1 a t+1 The Q-table is updated using a greedy strategy, where Q(s) is the action to be selected. t a t ) represents state s t Use action a t The value, where 'a' represents the soft update weight, and 'r' represents the value. t+1 Let γ represent the reward at time step t+1, and let γ represent the discount rate, where t = t+1.

[0045] S10, in each iteration, constantly judges the quality of the current population, and based on the trained Q table, uses the current population state of the chromosome to adaptively update the discrete particle swarm parameter ω, and selects the ω that is most beneficial to the chromosome of the current population.

[0046] S11, set a random number between [0, 1], compare the size of the random number with ω, and use the methods of cargo mutation and yard selection mutation to mutate the chromosome part of the cargo and the yard selection part of the vehicle respectively. Use a roulette wheel betting method to select one of the two mutations as the chromosome mutation operation.

[0047] S12, select the POX chromosome crossover method, and crossover the chromosomes of the yard section and the chromosomes of the picked goods respectively.

[0048] S13, select a two-point crossover method to crossover the chromosomes of the storage yard and the chromosomes of the picked goods respectively.

[0049] S14, perform chromosome update operation, evaluate and calculate the fitness value of each particle in the current population, and update the optimal P-position of the updated individual. i and the global optimal position P g .

[0050] S15, check if the iteration count meets the condition. If it does, stop the iteration and output the global optimal position P. g This is used as the optimal scheduling result.

[0051] The working principle of this invention is as follows: Based on the combination of the Sarsa algorithm and the QLearning algorithm in reinforcement learning, the key parameter ω of the discrete particle swarm optimization algorithm is adaptively selected. First, the entry and exit times of each vehicle and the target goods to be picked up by the vehicle are analyzed. K-means clustering is performed on the target goods. According to the stacking rules of the stacking yard, the historical dwell time of the vehicle in the stacking yard is analyzed to give the distribution of the picking time of the vehicle in each stacking yard. Second, a combination of global search, local search and random search is used to generate a high-quality initial population. Finally, ω in the discrete particle swarm optimization algorithm is intelligently updated through reinforcement learning. ω determines the diversity of particle search. When ω is large, the global search ability is strong and the local search ability is weak; when ω is small, the local search ability is strong and the global search ability is weak. The intelligent reinforcement learning algorithm is used to adaptively select the parameter ω to solve the vehicle stacking yard scheduling problem in an efficient way.

[0052] The above description is only used to illustrate the technical solution of the present invention and is not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention, as long as they do not depart from the spirit and scope of the technical solution of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A logistics park vehicle pickup scheduling method based on adaptive discrete particle swarm optimization algorithm, characterized by: The specific steps are as follows: S1 analyzes the operational data related to vehicles in the logistics park. Based on the entry and exit times of each vehicle and the target goods to be picked up, K-means clustering is performed on the target goods. Specifically, according to the yard mixing rules, the historical dwell time of vehicles in the yard is analyzed to give the distribution T of the picking time of vehicles in each yard. i ; S2, the yard allocation part is encoded as follows: the chromosome length of the yard allocation part is set to the sum of the cargo types of all vehicles picking up cargo, each gene position is represented by an integer, and arranged in order according to the vehicle and the cargo type of the vehicle. Each integer represents the yard number that meets the requirements for the current cargo picked up by the vehicle. S3, the picking goods portion is encoded, specifically: the chromosome length of the picking goods portion is set to the sum of the types of goods picked up by all vehicles, each gene is directly encoded with the vehicle number, and the order in which the vehicle numbers appear indicates the order in which the vehicle picked up the goods. S4, set up a global search to determine the initial population. Specifically, set up an array of length equal to the number of yards. The value in each array corresponds to the working time of the corresponding yard. Randomly select a vehicle i from the set of picking vehicles. Starting from the first picking goods of the current vehicle i, add the picking time of the yard that meets the target picking goods to the corresponding time in the array. Select the shortest time as the current target yard and update the array. Continue in this way until all vehicles have picked up their goods. S5, set up a local search to determine the initial population. Specifically, set up an array of length equal to the number of yards. The value in each array corresponds to the working time of the corresponding yard. Randomly select a vehicle i from the set of picking vehicles. Starting from the first picking goods of the current vehicle i, add the picking time of the yard that meets the target picking goods to the corresponding time in the array. Select the shortest time as the current target yard. When vehicle i is selected, reset the array to 0 and select the next vehicle until all vehicles have picked up their goods. S6, set up a random search to determine the initial population, specifically: randomly select a vehicle i from the set of pickup vehicles, randomly sort the order of the target pickup goods of vehicle i, and randomly select a yard from the set of yards that meet the requirements of the sorted target goods, and so on, until all vehicles have finished picking up their goods. S7, set the maximum number of algorithm iterations to iter; S8: Initialize the state set and action set in reinforcement learning, select actions using a greedy strategy, and calculate the reward based on the state and fitness value of the population chromosome at the current time. S9. The Sarsa algorithm updates the Q table and determines whether the current Sarsa update count is greater than the number of transformations. If it is greater, the QLearning algorithm is started to update the Q table, and a greedy strategy is used to select actions. S10, in each iteration, constantly judge the quality of the current population, and based on the trained Q table, use the current population state of the chromosome to adaptively update the discrete particle swarm parameter ω, and select the ω that is most beneficial to the chromosome of the current population. S11, Set a random number between [0,1], and compare the size of the random number with ω to adopt the method of cargo variation and yard selection variation; S12, Select the POX chromosome crossover method, and crossover the chromosomes of the yard section and the chromosomes of the picked goods respectively; S13, Select a two-point crossover method to crossover the chromosomes of the storage yard and the chromosomes of the picked goods respectively; S14, perform chromosome update operation, evaluate and calculate the fitness value of each particle in the current population, and update the optimal P-position of the updated individual. i and the global optimal position P g ; S15, check if the iteration count meets the condition. If it does, stop the iteration and output the global optimal position P. g This is used as the optimal scheduling result.

2. The logistics park vehicle pickup scheduling method based on adaptive discrete particle swarm optimization algorithm according to claim 1, characterized in that: The specific steps of S8 are as follows: 1) Initialize the state set in reinforcement learning, set the value range of the state set to [0.5,1], and divide the value range into 10 intervals as state1, state2...state10; 2) Initialize the action set in reinforcement learning, set the range of the state set to [0,1], and divide the range of values ​​into 10 intervals, which are action1, action2...action10; 3) Use a greedy strategy to select actions based on the state, and calculate the reward based on the state and fitness of the population chromosome at the current moment.

3. The logistics park vehicle pickup scheduling method based on adaptive discrete particle swarm optimization algorithm according to claim 1, characterized in that: The specific steps of S9 are as follows: 1) Set the number of algorithm transformations t = 1, and the total number of algorithm transformations T = 25; 2) The Sarsa algorithm is based on Q(s) t a t )=(1-a)Q(s t a t )+a(r t+1 +γQ(s t+1 a t+1 Update the Q table; 3) Determine if the current t is greater than the number of conversions. If it is, start S14; if it is less, start S12. 4) The QLearning algorithm is based on Q(s) t a t )=(1-a)Q(s t a t )+a(r t+1 +γmax a Q(s t+1 a t+1 Update the Q table and use a greedy strategy to select actions.

4. The logistics park vehicle pickup scheduling method based on adaptive discrete particle swarm optimization algorithm according to claim 1, characterized in that: Specifically, S11 involves: performing mutations on the chromosome portion of the picked goods and on the yard selection portion of the vehicle, respectively, and selecting one of the two mutations as the chromosome mutation operation using a roulette wheel betting method.

Images